What is the point in a Kalman filter ? Why don't people just come up with a decent estimator in the first place and cut out all the bullshit ?
Lack of computational power or lack of information. In both cases you can only approximate. Look at this. http://en.wikipedia.org/wiki/Kalman_filter
I am aware of wikipedia.
I refer you to the original question. You may assume infinite computational power.
Ok, then the second premise where not all the information is not there, or not to a level of accuracy. A simple example is the Radar, there is always interference with the signal. The filter approximates to find and enemy location, in a similar fashion to the Bayesian stats used in GPS.
Well, i have no idea what you mean by similar fashion to Bayesian stats used in GPS.
The Kalman filter is as highly dependant on the initial estimate as of the state variables.
Given poor level of initial conditions or information, it does not matter how efficient your filter is you are not going to get any convergence. ie. decent predicition, without the initial predictor being within a certain boundary.
If this is the case, then the Kalman filter is only really measuring parameters rather than predicting, and once these parameters are estimated within what you called a level of accuracy, then we are back to how good your initial prediction is, in deciding how effective the filter is.
This is my initial question.
ie. Why bother with the filter, and just concentrate on the estimator ?